基于数据相似度的间歇过程在线监控

On-line batch processes monitoring based on similarity of the process data

为了提高间歇过程批次之间产品的一致性,并及时发现过程中的异常情况,提出一种基于过程数据相似度的多变量统计监控方法对间歇过程的操作进行在线监控。该方法将正常批次轨迹与参考批次轨迹之间的相似度作为一种新的监控指标,并利用核密度方法估计相似度的概率密度函数,计算出控制限,在批次反应过程中利用Kalman滤波器对当前批次的数据进行实时的估计从而实现在线监控。该方法和传统多向主元分析方法的监控性能在一个青霉素发酵仿真系统上进行了比较。仿真结果表明:该方法检测出渐变型扰动比MPCA方法提前了30 h。

A multivariate statistical process monitoring method was developed based on a similarity analysis of the operating data for on-line batch process monitoring. The similarity between the entire trajectory of a normal batch run and a predescribed reference trajectory is used as the monitoring index. The control limits are computed using the kernel density function. The on-line monitoring uses a Kalman filter which can estimate the entire trajectory of the current batch run. Comparison of the monitoring performance of the method with that of the traditional muhiway principal component analysis （MPCA） method on a benchmark fed batch penicillin fermentation process shows that detecting the ramp faults using this method is 30 h faster than using the MPCA method.